Research map / Vision-language and multimodal / Human and embodied modeling
Solar Power Generation Forecasting via Multimodal Feature Fusion (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces the CBNU-SolarPower multimodal dataset (soiled/clean panel images, sky images, weather data) and a fusion model (FCN soiling features, CNN cloud features, concatenated into a DNN) for solar power generation forecasting.
Problem
Solar power generation forecasting is degraded by panel soiling and weather-related sunlight reduction, and prior work handles these factors separately and without multimodal data.
Approach
Collect a multimodal dataset of soiled panel images, sky images, and weather data, then fuse soiling (FCN segmentation), sunlight (CNN cloud classification), and weather features by concatenation into a DNN regressor for power forecasting.
Key results
- Weather+Soiling+Sunlight achieves R2 of 81.79%, RMSE 35.01, MAE 25.44.
- Adding soiling to weather raises R2 from 50.54% to 75.26%.
- Adding sunlight features further raises R2 by 6.53 points to 81.79%.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add soiling features to weather features (ablation) | Solar power forecasting model · CBNU-SolarPower test set | R2 (%): +24.72% | Helped |
| Add sunlight features to weather and soiling features (ablation) | Solar power forecasting model · CBNU-SolarPower test set | R2 (%): +6.53% | Helped |
Limitations
- Data collection is expensive, limiting dataset scale.
- Soiling segmentation accuracy is only 71.2% mIoU.
- Soiling was artificially created rather than naturally accumulated.
Experimental setup
Datasets: CBNU-SolarPower SWIMCAT
Benchmarks: CBNU-SolarPower
Metrics: R2 RMSE MAE mIoU
Baselines: DeepSolar-Eye SUNSET
Builds on: Fully Convolutional Network SUNSET DeepSolar-Eye
Closest papers by meaning
- SolarCube: An Integrative Benchmark Dataset Harnessing Satellite and In-situ Observations for Large-scale Solar Energy Forecasting (NeurIPS 2024 · similarity 0.82)
- Terra: A Multimodal Spatio-Temporal Dataset Spanning the Earth (NeurIPS 2024 · similarity 0.80)
- EcoMapper: Generative Modeling for Climate-Aware Satellite Imagery (ICML 2025 · similarity 0.79)
- Time Weaver: A Conditional Time Series Generation Model (ICML 2024 · similarity 0.77)
- Causally Aware Generative Adversarial Networks for Light Pollution Control (AAAI 2024 · similarity 0.77)
- MMM-RS: A Multi-modal, Multi-GSD, Multi-scene Remote Sensing Dataset and Benchmark for Text-to-Image Generation (NeurIPS 2024 · similarity 0.77)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.